Research map / Graph learning / Graph algorithms and clustering
ProGraML: A Graph-based Program Representation for Data Flow Analysis and Compiler Optimizations
Chris Cummins, Zacharias V. Fisches, Tal Ben-Nun, Torsten Hoefler, Michael F P O’Boyle, Hugh Leather
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add data-flow edges to graph representation (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.38 -> 7.76 when removed (+129.6%) | Improved |
| add backward edges (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.38 -> 4.16 when removed | Improved |
| add control edges (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.38 -> 3.88 when removed | Improved |
| add call edges (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.38 -> 3.88 when removed | Improved |
| add PROGRAML node vocabulary (vs no vocab) (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.70 -> 3.38 | Improved |
| replace PROGRAML vocabulary with inst2vec vocabulary (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.38 -> 3.78 | Hurt |
| add edge position attributes (ablation) | PROGRAML on algorithm classification · Algorithm classification | error rate (%): 3.43 -> 3.38 when removed (+1.5%) | Improved |
| use PROGRAML graph representation with GGNN instead of CDFG or inst2vec | Heterogeneous device mapping · AMD and NVIDIA device mapping | error rate (%): AMD 19.7 (inst2vec) -> 13.4; NVIDIA 21.5 -> 20.0 | Improved |
| use PROGRAML instead of XFG for algorithm classification | Algorithm classification · Algorithm classification | error rate (%): 4.29 -> 3.38 | Improved |
| use PROGRAML graph model instead of CDFG on data flow analysis tasks | Subexpressions and Reachability/Dominance analyses · DDF-30 | F1: Subexpressions 0.009 (CDFG) -> 0.996; Reachability and Dominance roughly equal | Improved |
| scale inference to full DDF test set with T=200 steps (vs T=30 training) | PROGRAML · DDF full test set | F1: DDF-30 Dominance 0.991 -> DDF 0.123; Reachability 0.997 -> 0.943 | Hurt |
Later papers that compared with this method
1 later accepted paper reports ProGraML as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- PERFOGRAPH: A Numerical Aware Program Graph Representation for Performance Optimization and Program Analysis (NeurIPS 2023 · better on 5 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PERFOGRAPH: A Numerical Aware Program Graph Representation for Performance Optimization and Program Analysis (NeurIPS 2023 · similarity 0.89)
- Behavioral Embeddings of Programs: A Quasi-Dynamic Approach for Optimization Prediction (ICLR 2026 · similarity 0.88)
- LLMDFA: Analyzing Dataflow in Code with Large Language Models (NeurIPS 2024 · similarity 0.86)
- ComBack: A Versatile Dataset for Enhancing Compiler Backend Development Efficiency (NeurIPS 2024 · similarity 0.85)
- ReGAL: Refactoring Programs to Discover Generalizable Abstractions (ICML 2024 · similarity 0.85)
- Chain of Execution Supervision Promotes General Reasoning in Large Language Models (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
